Smooth toric Deligne-Mumford stacks Barbara Fantechi, Etienne Mann, Fabio Nironi
To cite this version:
Barbara Fantechi, Etienne Mann, Fabio Nironi. Smooth toric Deligne-Mumford stacks. Jour- nal für die reine und angewandte Mathematik, Walter de Gruyter, 2010, 648, pp.201-244. 10.1515/CRELLE.2010.084. hal-00166751v2
HAL Id: hal-00166751 https://hal.archives-ouvertes.fr/hal-00166751v2 Submitted on 14 Sep 2012
HAL is a multi-disciplinary open access L’archive ouverte pluridisciplinaire HAL, est archive for the deposit and dissemination of sci- destinée au dépôt et à la diffusion de documents entific research documents, whether they are pub- scientifiques de niveau recherche, publiés ou non, lished or not. The documents may come from émanant des établissements d’enseignement et de teaching and research institutions in France or recherche français ou étrangers, des laboratoires abroad, or from public or private research centers. publics ou privés. AUTHOR’S COPY | AUTORENEXEMPLAR
J. reine angew. Math. 648 (2010), 201—244 Journal fu¨r die reine und DOI 10.1515/CRELLE.2010.084 angewandte Mathematik ( Walter de Gruyter Berlin New York 2010
Smooth toric Deligne-Mumford stacks
By Barbara Fantechi at Trieste, Etienne Mann at Montpellier and Fabio Nironi at New York
Abstract. We give a geometric definition of smooth toric Deligne-Mumford stacks using the action of a ‘‘torus . We show that our definition is equivalent to the one of Borisov, Chen and Smith in terms of stacky fans. In particular, we give a geometric inter- pretation of the combinatorial data contained in a stacky fan. We also give a bottom up classification in terms of simplicial toric varieties and fiber products of root stacks.
Contents
Introduction 1. Notations and background 1.1. Conventions and notations 1.2. Smooth Deligne-Mumford stacks and orbifolds 1.3. Root constructions 1.4. Rigidification 1.5. Diagonalizable group schemes 1.6. Toric varieties 1.7. Picard stacks and action of a Picard stack 2. Deligne-Mumford tori 3. Definition of toric Deligne-Mumford stacks 4. Canonical toric Deligne-Mumford stacks 4.1. Canonical smooth Deligne-Mumford stacks 4.2. The canonical stack of a simplicial toric variety 5. Toric orbifolds 6. Toric Deligne-Mumford stacks 6.1. Gerbes and root constructions 6.2. Gerbes on toric orbifolds 6.3. Characterization of a toric Deligne-Mumford stack as a gerbe over its rigidification 7. Toric Deligne-Mumford stacks versus stacky fans 7.1. Toric Deligne-Mumford stacks as global quotients 7.2. Toric Deligne-Mumford stacks and stacky fans 7.3. Examples
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
202 Fantechi, Mann and Nironi, Deligne-Mumford stacks
Appendix A. Uniqueness of morphisms to separated stacks Appendix B. Action of a Picard stack Appendix C. Stacky version of Zariski s Main Theorem References
Introduction
A toric variety is a normal, separated variety X with an open embedding T , X of a torus such that the action of the torus on itself extends to an action on X. To a toric! variety one can associate a fan, a collection of cones in the lattice of one-parameter subgroups of T. Toric varieties are very important in algebraic geometry, since algebro-geometric prop- erties of a toric variety translate in combinatorial properties of the fan, allowing to test con- jectures and produce interesting examples.
In [10] Borisov, Chen and Smith define toric Deligne-Mumford stacks as explicit global quotient (smooth) stacks, associated to combinatorial data called stacky fans. Later, Iwanari proposed in [22] a definition of toric triple as an orbifold with a torus action having a dense orbit isomorphic to the torus1) and he proved that the 2-category of toric triples is equivalent to the 2-category of ‘‘toric stacks (We refer to [22] for the definition of ‘‘toric stacks ). Nevertheless, it is clear that not all toric Deligne-Mumford stacks are toric triples, since some of them are not orbifolds.
Then the generalization of the D-collections defined for toric varieties by Cox in [14] was done by Iwanari in [23] in the orbifold case and by Perroni in [31] in the general case.
In this paper, we define a Deligne-Mumford torus T as a Picard stack isomorphic to T BG, where T is a torus, and G is a finite abelian group; we then define a smooth toric Deligne-Mumford stack as a smooth separated Deligne-Mumford stack with the action of a Deligne-Mumford torus T having an open dense orbit isomorphic to T. We prove a classification theorem for smooth toric Deligne-Mumford stacks and show that they coin- cide with those defined by [10].
The first main result of this paper is a bottom-up description of smooth toric Deligne- Mumford stacks, as follows: the structure morphism X X to the coarse moduli space factors canonically via the toric morphisms !
X Xrig Xcan X ! ! ! where X Xrig is an abelian gerbe over Xrig; Xrig Xcan is a fibered product of roots of toric divisors;! and Xcan X is the minimal orbifold! having X as coarse moduli space. Here X is a simplicial toric! variety, and Xrig and Xcan are smooth toric Deligne-Mumford stacks. More precisely, this bottom up construction can be stated as follows.
Theorem I. Let X be a smooth toric Deligne-Mumford stack with Deligne-Mumford torus isomorphic to T BG. Denote by X the coarse moduli space of X. Denote by n the number of rays of the fan of X.
1) For the meaning of orbifold in this paper, see §1.2.
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 203
n rig (1) There exist unique a1; ...; an A N>0 such that the stack X is isomorphic, as toric Deligne-Mumford stack , to
a1 an can Xcan can can can Xcan D1 = X X Dn = ; qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi can where Di is the divisor corresponding to the ray ri.
l rig ... l X X (2) Given G mbj . There exist L1; ; L in Pic such that is isomorphic, as j 1 Q toric Deligne-Mumford stack, to
b1 rig bl rig L1=X X rig X rig Ll=X : qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi qffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi rig rig Moreover, for any j A 1; ...; l , the class Lj in Pic X =bj Pic X is unique. f g In the process, we get a description of the Picard group of smooth toric Deligne- Mumford stacks, which allows us to characterize weighted projective stacks as complete toric orbifolds with cyclic Picard group (cf. Proposition 7.28). Moreover, we classify all complete toric orbifolds of dimension 1 (cf. Example 7.31). We also show that the natural map from the Brauer group of a smooth toric Deligne-Mumford stack with trivial generic stabilizer to its open dense torus is injective (cf. Theorem 6.11).
The second main result of this article is to give an explicit relation between the smooth toric Deligne-Mumford stacks and the stacky fans.
Theorem II. Let X be a smooth toric Deligne-Mumford stack with coarse moduli space the toric variety denoted by X. Let S be a fan of X in NQ : N nZ Q. Assume that the rays of S generate NQ. There exists a stacky fan such that X is isomorphic, as toric Deligne-Mumford stack, to the smooth Deligne-Mumford stack associated to the stacky fan. Moreover, if X has a trivial generic stabilizer then the stacky fan is unique.
When the smooth toric Deligne-Mumford stack X has a generic stabilizer the non- uniqueness of the stacky fan comes from three di¤erent choices. We refer to Remark 7.26 for a more precise statement. This result gives a geometrical interpretation of the combina- torial data of the stacky fan. In fact, the stacky fan can be read o¤ the geometry of the smooth toric Deligne-Mumford stack just like the fan can be read o¤ the geometry of the toric variety. Notice that one can deduce the above theorem when X is an orbifold from [31], Theorem 2.5, and [23], Theorem 1.4, and the geometric characterization of [24], Theorem 1.3.
In the first part of this article, we fix the conventions and collect some results on smooth Deligne-Mumford stacks, root constructions, rigidification, toric varieties, Picard stacks and the action of a Picard stack. In Section 2, we define Deligne-Mumford tori. Sec- tion 3 contains the definition of smooth toric Deligne-Mumford stacks. In Section 4, we first define canonical smooth Deligne-Mumford stacks and then we show that the canonical stack associated to a simplicial toric variety is a smooth toric Deligne-Mumford stack (cf. Theorem 4.11). In Section 5, we prove the first part of Theorem I. In Section 6, we first prove in Proposition 6.9 that the essentially trivial banded gerbes over X are in bijection
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
204 Fantechi, Mann and Nironi, Deligne-Mumford stacks with finite extensions of the Picard group of X; then, we show that the natural map from the Brauer group of a smooth toric Deligne-Mumford stack with trivial generic stabilizer to its open dense torus is injective (cf. Theorem 6.11). Finally, we prove the second statement of Theorem I. In Section 7, we prove Theorem II and give some explicit examples. In Appendix B, we have put some details about the action of a Picard stack.
Acknowledgments. The authors would like to acknowledge support from IHP, Mittag-Le¿er Institut, SNS where part of this work was carried out, as well as the Euro- pean projects MISGAM and ENIGMA. We would like to thank Ettore Aldrovandi, Lev Borisov, Jean-Louis Colliot-The´le`ne, Andrew Kresch, Fabio Perroni, Ilya Tyomkin and Angelo Vistoli for helpful discussions; in particular Aldrovandi for explanations about group-stacks and reference [11], Borisov for pointing out a mistake in a preliminary ver- sion, Colliot-The´le`ne for [20], §6, Tyomkin for [9] and Vistoli for useful information about the classification of gerbes.
1. Notations and background
1.1. Conventions and notations. A scheme will be a separated scheme of finite type over C. A variety will be a reduced, irreducible scheme. A point will be a C-valued point. The smooth locus of a variety X will be denoted by Xsm.
We work in the e´tale topology. For an algebraic stack X, we will write that x is a point of X or just x A X to mean that x is an object in X C ; we denote by Aut x the auto- morphism group of the point x. We will say that a morphism between stacks is unique if it is unique up to a unique 2-arrow. As usual, we denote Gm the sheaf of invertible sections in OX on the e´tale site of X.
1.2. Smooth Deligne-Mumford stacks and orbifolds. A Deligne-Mumford stack will be a separated Deligne-Mumford stack of finite type over C; we will always assume that its coarse moduli space is a scheme. An orbifold will be a smooth Deligne-Mumford stack with trivial generic stabilizer. For a smooth Deligne-Mumford stack X, we denote by eX or just e the natural morphism from X to its coarse moduli space X, which is a variety with finite quotient singularities.
Let i : U X be an open embedding of irreducible smooth Deligne-Mumford stacks with complement! of codimension at least 2. We have that:
The natural map i : Pic X Pic U is an isomorphism. ! For any line bundle L A Pic X , the natural morphism i : H 0 X; L H 0 U; i L is also an isomorphism. !
The inertia stack, denoted by I X , is defined to be the fibered product I X : X X X X. A point of I X is a pair x; g with x A X and g A Aut x . The inertia stack of a smooth Deligne-Mumford stack is smooth but di¤erent components will in general have di¤erent dimensions. The natural morphism I X X is representable, unramified, proper and a relative group scheme. The identity section ! gives an irreducible component canonically isomorphic to X; all other components are called twisted sectors.
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 205
A smooth Deligne-Mumford stack of dimension d is an orbifold if and only if all the twisted sectors have dimension e d 1, and is canonical if and only if all twisted sectors have dimension e d 2. Remark 1.1 (sheaves on global quotients). According to [33], Appendix, a coherent sheaf on a Deligne-Mumford stack Z=G is a G-equivariant sheaf on Z, i.e., the data of a coherent sheaf LZ on Z and for every g A G an isomorphism jg : LZ g LZ such that j h j j . ! gh g h Notice that if Z is a subvariety of Cn of codimension higher or equal than two then an invertible sheaf on Z=G is the structure sheaf OZ and a one dimensional representa- tion of G, i.e., w : G C . A global section of such an invertible sheaf on Z=G is a w- ! equivariant global section of OZ.
We end this subsection with a proposition extending to stacks a property of separated schemes. We will prove it in Appendix A.
Proposition 1.2. Let X and Y be two Deligne-Mumford stacks. Assume that X is normal and Y is separated. Let i : U , X be a dominant open immersion of the Deligne- Mumford stack U.IfF; G : X Y are! two morphisms of stacks such that there exits a b ! 2-arrow F i G i then there exists a unique 2-arrow a : F G such that a idi b. ) ) The previous proposition is well known for X a reduced scheme and Y a separated scheme. Nevertheless, if X is not a normal stack we have the following counter-example: Y B X X Y Let be m2. Let be a rational curve with one node. Let F1 : (resp. F2) be a stack morphism given by a non-trivial (resp. trivial) double cover of X.! Putting U X node , the proposition is false. nf g
1.3. Root constructions. For this subsection we refer to the paper of Cadman [12] (see also [2], Appendix B). In this part X will be a Deligne-Mumford stack over C (it is enough to assume that X is Artin.)
1.3.a. Root of an invertible sheaf. This part follows closely [2], Appendix B. Let L be an invertible sheaf on the Deligne-Mumford stack X. Let b be a positive integer. We denote by b L=X the following fiber product pffiffiffiffiffiffiffiffiffiffi b L=X BC ! pffiffiffiffiffiffiffiffiffiffi r 5b ? ? ? L ? X? B?C y ! y nb where 5b : BC BC sends an invertible sheaf M over a scheme S to M . More ex- plicitly, an object! of b L=X over f : S X is a couple M; j where M is an invertible nb @! sheaf M on the schemepffiffiffiffiffiffiffiffiffiffiS and j : M f L is an isomorphism. The arrows are defined in an obvious way. !
b 1=b The morphism L=X BC corresponds to an invertible sheaf, denoted by L in b ! [8], on L=X whosepb-thffiffiffiffiffiffiffiffiffiffi power is isomorphic to the pullback of L. pffiffiffiffiffiffiffiffiffiffi
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
206 Fantechi, Mann and Nironi, Deligne-Mumford stacks
b X X The stack L= is a mb-banded gerbe over (see the second paragraph of Subsec- tion 6.1 below).p Theffiffiffiffiffiffiffiffiffiffi Kummer exact sequence
5b 1 m Gm Gm 1 ! b ! ! ! 1 X G 2 X induces the boundary morphism q : Het´ ; m Het´ ; mb . The cohomology class of the b 2 ! 1 m -banded gerbe L=X in H X; m is the image by q of the class L A H X; Gm . b et´ b et´ pffiffiffiffiffiffiffiffiffiffi The gerbe is trivial if and only if the invertible sheaf L has a b-th root in Pic X . More b X b X generally, the gerbe L= is isomorphic, as a mb-banded gerbe, to L0= if and only if L L in Pic X =pb PicffiffiffiffiffiffiffiffiffiffiX . We have the following morphism of shortpffiffiffiffiffiffiffiffiffiffiffiffi exact sequences: 0 b 0 Z Z Z=bZ 0 ! ! ! ! 1:3 ? ? ? ? 0 Pic?X Pic b?L=X Z= bZ 0 ! y ! y ! ! pffiffiffiffiffiffiffiffiffiffi where the first and second vertical morphisms are defined by 1 L and 1 L1=b, respec- tively. 7! 7!
1.3.b. Roots of e¤ective Cartier divisors. In the articles [12] and [2], the authors de- fine the notion of root of an invertible sheaf with a section on an algebraic stack: here, we only consider roots of e¤ective Cartier divisors on a smooth algebraic stack, since this is what we will use.
n n Let n be a positive integer. Consider the quotient stack A = C where the action n n n of C is given multiplication coordinates by coordinates. Notice that A = C is the n moduli stack of n line bundles with n global sections. Let a : a1; ...; an A N>0 be an n n n n n-tuple. Denote by 5a : A = C A = C the stack morphism defined by sending ai ai ! n n xi x and li l where xi (resp. li) are coordinates of A (resp. C ). 7! i 7! i
Let X be a smooth algebraic stack. Let D : D1; ...; Dn be n e¤ective Cartier divi- sors. The a-th root of X; D is the fiber product a n n D=X A = C ! pffiffiffiffiffiffiffiffiffiffiffi p r 5a ? ? ? D n ? n X? A =?C : y ! y a n n The morphism D=X A = C corresponds to the e¤ective Cartier divisors DD~ ~ ~ ! ~ 1 : DD1; ...; DDn p, whereffiffiffiffiffiffiffiffiffiffiffi DDi is the reduced closed substack p Di red. More explicitly, an a ~ ~ object of D=X over a scheme S is a couple f ; DD1; ...; DDn where f : S X is a mor- ~ ! phism andp forffiffiffiffiffiffiffiffiffiffiffi any i, Di is an e¤ective divisor on S such that a iDDi f Di. We have the following properties:
a ai (1) The fiber product of Di=X over X is isomorphic to D=X (cf. [12], Remark 2.2.5). pffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffi
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 207
a (2) The canonical morphism D=X X is an isomorphism over X Di. ! n pffiffiffiffiffiffiffiffiffiffiffi Si
(3) If X is smooth, each Di is smooth and the Di have simple normal crossing then a D=X is smooth (cf. Section 2.1 of [8]) and DD~i have simple normal crossing. pffiffiffiffiffiffiffiffiffiffiffi (4) We have the following morphism of short exact sequences (cf. [12], Corollary 3.1.2)
n n a n 0 Z Z Z=aiZ 0 ! ! ! i 1 ! Q 1:4 ? ? ? ? n ?X p a?D X q 0 Picy Pic y = Z =aiZ 0 ! ! ! i 1 ! pffiffiffiffiffiffiffiffiffiffiffi Q where the first and second vertical morphisms are defined by ei O Di and ei O DD~i , respectively. Every invertible sheaf L A Pic a D=X can be written7! in a unique7! way as n ~ pffiffiffiffiffiffiffiffiffiffiffi L G p M n O kiDDi where M A Pic X and 0 e ki < ai; the morphism q maps L to i 1 k1; ...; kn . Q
We finish this section with the following observation. Let D1 and D2 be two e¤ective a a; a Cartier divisors on X such that D X D 3j. The stacks D W D =X and D ; D =X 1 2 1 2 1 2 are not isomorphic. Indeed, the stabilizer group at any pointpffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi in the preimage ofpxffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiA D1 X D2 a a; a in D W D =X (resp. D ; D =X)ism (resp. m m ). 1 2 1 2 a a a pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi pffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffiffi 1.4. Rigidification. In this subsection, we sum up some results on the rigidification of an irreducible d-dimensional smooth Deligne-Mumford stack X. Intuitively, the rigidifi- cation of X by a central subgroup G of the generic stabilizer is constructed as follows: first, one constructs a prestack where the objects are the same and the automorphism groups of each object x are the quotient AutX x =G; then the rigidification X( G is the stackification of this prestack. For the most general construction we refer to [3], Appendix A (see also [1], Section 5.1, [32] and [2], Appendix C).
We consider the union I gen X H I X of all d-dimensional components of I X ;itis a subsheaf of groups of I X over X which is called the generic stabilizer. Most of the time in this article, we will rigidify by the generic stabilizer. In this case, we write Xrig in order to mean X( I gen X and call it the rigidification. The rigidification r : X Xrig has the following properties: ! (1) The coarse moduli space of Xrig is the coarse moduli space of X.
(2) Xrig is an orbifold.
(3) If X is an orbifold then Xrig is X.
(4) The morphism r makes X into a gerbe over Xrig.
We refer to [1], Theorem 5.1.5(2), for the proof of the following proposition.
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
208 Fantechi, Mann and Nironi, Deligne-Mumford stacks
Proposition 1.5 (universal property of the rigidification). Let X be a smooth Deligne- Mumford stack. Let Y be an orbifold. Let f : X Y be a dominant stack morphism. Then there exists g : Xrig Y and a 2-morphism a :!g r f such that the following is 2-
commutative: ! )
X r Xrig !
bg f ? ? Y?: y If there exists g : Xrig Y and a 2-morphism a : g r f satisfying the same property 0 ! 0 0 ) then there exists a unique g : g g such that a g idr a . 0 ) 0 1.5. Diagonalizable group schemes. In this short subsection, we recall some results on diagonalizable groups.
Definition 1.6. A group scheme G over Spec C will be called diagonalizable if it is isomorphic to the product of a torus and a finite abelian group.
We use multiplicative notation for diagonalizable group. For any diagonalizable 4 group G, its character group G : Hom G; C is a finitely generated abelian group (or coherent Z-module). The duality contravariant functor G G4 induces an equivalence of categories from diagonalizable to coherent Z-module. Its7! inverse functor is given by 4 F GF : Hom F; C . Both G G and F GF are contravariant and exact. 7! 7! 7! 1.6. Toric varieties. We recall some results on toric varieties that can be found in [17] (see also [15]). The principal construction used in this paper is the description of toric varieties as global quotients found by Cox (see [13]).
We fix a torus T, and denote by M T4 the lattice of characters and by N : Hom M; Z the lattice of one-parameter subgroups. A toric variety X with torus T corresponds to a fan S X , or just S,inNQ : N nZ Q, which we will always assume to be simplicial.
S Let r1; ...; rn be the one-dimensional cones, called rays, of . For any ray ri, denote by vi the unique generator of r X N. For any i in 1; ...; n , we denote by Di the irreduc- i f g ible T-invariant Weil divisor defined by the ray ri. The free abelian group of T-invariant Weil divisor is denoted by L. n Let i : M L be the morphism that sends m m vi . If the rays span NQ (which ! 7! i 1 P is not a strong assumption2)), the morphism i is injective, and fits into an exact sequence in Coh Z i 1:7 0 M L A 0; ! ! ! !
2) Indeed, if the rays do not span NQ then X is isomorphic to the product of a torus and a toric variety XX~ whose rays span NN~Q.
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 209 where A is the class group of X (i.e., the Chow group A1 X ). We deduce that the short exact sequence of diagonalizable groups
1:8 1 GA GL T 1: ! ! ! ! n n Let ZS H C be the GL C -invariant open subset defined as ZS : Zs, where S sSA Zs : x xi 3 0ifri B s . The induced action of GA on ZS has finite stabilizers (by the f j g n simpliciality assumption) and X is the geometric quotient ZS=GA, with torus C =GA (see [13], Theorem 2.1). For any i A 1; ...; n , the T-invariant Weil divisor Di H X is the geometric quotient f g
1:9 xi 0 X ZS =GA: f g
If X is smooth then the natural morphism L Pic X given by ei OX Di is sur- jective and has kernel M; in other words, it induces! a natural isomorphism7!A Pic X . ! If X is a d-dimensional toric variety, we will write X 0 for the union of the orbits of dimension f d 1; in other words, X 0 is the toric variety associated to the fan 0 Se1 : s A S dim s e 1 . The toric variety X is always smooth and the toric divisors 0 f j g Dr are smooth, disjoint, and homogeneous under the T-action (with stabilizer the one- dimensional subgroup which is the image of r).
1.7. Picard stacks and action of a Picard stack. Deligne defined Picard stacks in [7], Expose´ XVIII, as stacks analogous to sheaves of abelian groups. For the reader s conve- nience, we collect here a sketch of the definition and the main properties we need; details can be found in [7], Expose´ XVIII, and also in [26], Section 14.
Here we summarize the definition of a Picard stack. For the details we refer to Defi- nition B.1.
Definition 1.10. Let G be a stack over a base scheme S.APicard stack G over S is given by the following set of data:
a multiplication stack morphism m : G G G, also denoted by ! m g ; g g g ; 1 2 1 2 an associativity 2-arrow g g g g g g ; 1 2 3 ) 1 2 3 a commutativity 2-arrow g g g g . 1 2 ) 2 1 These data satisfy some compatibility relations, which we list in B.1.
The definition implies that there also exists an identity e : S G and an inverse i : G G with the obvious properties; in particular, a 2-arrow e : e !g g. ! )
Definition 1.11 (see [7], Section 1.4.6). Let G, G0 be two Picard stacks. A morphism of Picard stacks F : G G0 is a morphism of stacks and a 2-arrow a such that for any two ! objects g1, g2 in G, we have a F g g F g F g : 1 2 ) 1 2
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
210 Fantechi, Mann and Nironi, Deligne-Mumford stacks
Again we refer to Appendix B for the list of compatibilities satisfied by a. The Picard stacks over S form a category where the objects are Picard stacks and morphisms are equivalence classes of morphisms of Picard stacks.
Remark 1.12. To any complex G : G 1 G0 of sheaves of abelian groups, we ! can associate a Picard stack G. In this paper, G will be a complex of diagonalizable groups 1 0 and the associated Picard stack is the quotient stack G =G . Proposition 1.13 (see [7], Proposition 1.4.15). The functor that associates to a length 1 complex of sheaves of abelian groups a Picard stack induces an equivalence of categories 1; 0 between the derived category, denoted by D S; Z , of length 1 complexes of sheaves of abelian groups and the category of Picard stacks.
In particular, if G is any sheaf of abelian groups on the base scheme S, the quotient S=G , i.e. the gerbe BG, is naturally a Picard stack. We finish this section with a sketch of the definition of an action of a Picard stack on a stack. This is a generalization of the action of a group scheme on a stack defined by Romagny in [32]. We refer to Definition B.12 for the details.
Definition 1.14 (action of a Picard stack). Let G be a Picard stack. Denote by e the neutral section and by the corresponding 2-arrow. Let X be a stack. An action of G on X is the following data:
a stack morphism a : G X X, also denoted by a g; x g x; ! a 2-arrow e x x; ) an associativity 2-arrow g g x g g x . 1 2 ) 1 2 These data satisfy some compatibility relations, which we list in Appendix B.
2. Deligne-Mumford tori
In this section we define Deligne-Mumford tori which will play the role of the torus for a toric variety.
We start with a technical lemma.
Lemma 2.1. Let f : A0 A1 be a morphism of finitely generated abelian groups such that ker f is free. In the derived! category of complexes of finitely generated abelian groups 0 of length 1, the complex A0 A1 is isomorphic to ker f coker f . ! ! Proof. We have a morphism of complexes
f ff~ A0 A1 A0=A0 A1=A0 ! ! tor ! tor induced by the quotient morphisms. As ker f is free, we deduce after a diagram chasing that this morphism is a quasi-isomorphism of complexes. In the derived category, we replace
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 211
1 0 Zl Q Zd l A =Ator with a projective resolution . Then the mapping cone of the morphism 0 0 Zl !Zd l ~ 0 0 Zl Zd l of complexes 0 A =Ator Q : is ffQ : A =Ator which is ! ! ff~ ! ! quasi-isomorphic to A0=A0 A1=A0 . A morphism of free abelian groups f is quasi- tor 0 tor isomorphic to the complex ker!f coker f and this finishes the proof. r ! The reader who is familiar with the article [10] has probably recognized part of the construction of the stack associated to a stacky fan.
Remark 2.2. Let f : A0 A1 be a morphism of finitely generated abelian groups as ! in the above lemma. Applying the contravariant functor Hom ; C of Section 1.5 to the G 0 1 f complex A A , we get a length 1 complex of diagonalizable groups G 1 G 0 . Ac- ! A ! A cording to Remark 1.12, the associated Picard stack GA0 =GA1 is a Deligne-Mumford stack if and only if the cokernel of f is finite.
n 1 Example 2.3. Let w0; ...; wn be in N>0. Let f : Z Z that sends a0; ...; an to n ! wiai. We have that ker f Z and coker f Z=dZ where d : gcd w0; ...; wn . Hence, n B Pthe associated Picard stack is C md . Definition 2.4. A Deligne-Mumford torus is a Picard stack over Spec C which is 0 1 obtained as a quotient GA0 =GA1 , where f : A A is a morphism of finitely generated abelian groups such that ker f is free and coker f!is finite.
Let G be a finite abelian group. Notice that BG is a Deligne-Mumford torus. Recall that by Proposition 1.13, T BG has a natural structure of Picard stack. Definition 2.5. A short exact sequence of Picard S-stacks is the sequence of mor- phisms of Picard S-stacks associated to a distinguished triangle in D 1; 0 S . Proposition 2.6. Any Deligne-Mumford torus T is isomorphic as Picard stack to T BG where T is a torus and G is a finite abelian group. 0 1 Proof. Let T G 0 =G 1 with f : A A as above. The distinguished triangle G A A f ! 1; 0 ker Gf 0 G 1 G 0 0 coker Gf in the derived category D Spec C in- ! ! A ! A ! ! duces an exact sequence of Picard stacks 1 BG T T 1 where T : GA0 =GA1 . Proposition 1.13 and Lemma 2.1 imply that there! is a! non-canonical! ! isomorphism of Picard stacks T 1 BG T. r Note that the scheme T in the previous proof is the coarse moduli space of T.
3. Definition of toric Deligne-Mumford stacks
Definition 3.1. A smooth toric Deligne-Mumford stack is a smooth separated Deligne-Mumford stack X together with an open immersion of a Deligne-Mumford torus i : T , X with dense image such that the action of T on itself extends to an action a : T !X X. ! As in this paper all toric Deligne-Mumford stacks are smooth, we will write toric Deligne-Mumford stack instead of smooth toric Deligne-Mumford stack. We will see later
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
212 Fantechi, Mann and Nironi, Deligne-Mumford stacks in Theorem 7.24 that our definition a posteriori coincides with that in [10] via stacky fans. It seems natural to define a toric Deligne-Mumford stack by replacing smooth with normal in the above definition. All the definitions and results in this section apply also in this case, with the exception of Proposition 3.6 and Lemma 3.8. Ilya Tyomkin is currently working on this. A toric orbifold is a toric Deligne-Mumford stack with generically trivial stabilizer. A toric Deligne-Mumford stack is a toric orbifold if and only if its Deligne-Mumford torus is an ordinary torus. Hence, the notion of toric orbifold is the same as the one used in [22], Theorem 1.3.
Remark 3.2. (1) Separatedness of X and Proposition 1.2 imply that the action of T on X is uniquely determined by i.
(2) Notice that we have assumed in Section 1.2 that the coarse moduli space is a scheme. Without this assumption, if the coarse moduli space X of a toric Deligne-Mumford stack is a smooth and complete algebraic space then the main theorem of Bialynicki-Birula in [9] implies that X is a scheme. We don t know whether such an assumption is necessary in general.
(3) A toric variety admits a structure of toric Deligne-Mumford stack if and only if it is smooth.
Proposition 3.3. Let X be a smooth Deligne-Mumford stack together with an open dense immersion of a Deligne-Mumford torus i : T , X such that the action of T on itself extends to a stack morphism a : T X X. Then! the stack morphism a induces naturally an action of T on X. !
Proof. We will define a 2-arrow h : a e; idX idX and a 2-arrow )
s : a m; idX a idX; a ) such that they verify conditions (1) and (2) of Definition B.12. We will only prove the existence of h because the existence of s and the relations (1) and (2) follow with a similar argument.
Denote by e : Spec C T the neutral element of T and by m : T T T the ! ! multiplication on T. Denote by e the 2-arrow m e; idT idT. As the stack mor- ) phism a extends m, we have a 2-arrow a : a idT; i i m. Denote by b the 2-arrow ) e; idX i idT; i e; idT . Consider the two stack morphisms: )
idX TXXi :
a e; idX Applying Proposition 1.2 with the composition of the following 2-arrows
id b a id e; idT id e a i a e; idX i a idT; i e; idT i m e; idT i idT idX i; ) ) ) we deduce the existence of h : a e; idX idX. r )
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 213
Definition 3.4. Let X (resp. X0) be a toric Deligne-Mumford stack with Deligne- Mumford torus T (resp. T0). A morphism of toric Deligne-Mumford stacks F : X X0 is ! a morphism of stacks between X and X0 which extends a morphism of Deligne-Mumford tori T T0: ! Remark 3.5. The extended morphism F in the previous definition is unique by Pro- position 1.2. Moreover the definition of morphism between Picard stacks and Proposition 1.2 provide us the following 2-cartesian diagram:
F; F T X T j X0 T0 ! a r a ? ? 0 ? F ? X? X?0: y ! y Proposition 3.6. Let X be a toric Deligne-Mumford stack with Deligne-Mumford torus T. Let X (resp. T) be the coarse moduli space of X (resp. T). Then X has a structure of simplicial toric variety with torus T where the open dense immersion i : T , X and the action a : T X X is induced respectively by i : T , X and a : T X !X. ! ! ! Proof. The morphisms i and a induce morphisms on the coarse moduli spaces i : T X and a : T X X, by the universal property of the coarse moduli space. It is immediate! to verify that i!is an open embedding with dense image and a is an action, ex- tending the action of T on itself. On the other hand, since X is the coarse moduli space of X, it is a normal separated variety with finite quotient singularities. Therefore X is a toric variety, and it is simplicial by [21], §7.6, p. 121 (see also [15], Theorem 3.1, p. 28). r
Remark 3.7 (divisor multiplicities). According to [26], Corollary 5.6.1, the structure morphism e : X X induces a bijection on reduced closed substacks. For each ! 1 i 1; ...; n, denote by Di H X the reduced closed substack with support e Di . Since Di X Xsm is a Cartier divisor, there exists a unique positive integer ai such that 1 1 e Di X X ai Di X e X . We call a a ; ...; an the divisor multiplicities of X. sm sm 1 Let X be a toric Deligne-Mumford stack with Deligne-Mumford torus T T BG. By Appendix B, we have that BG acts on X. Proposition B.15 implies that we have an e´tale morphism j : G X I gen X . ! Lemma 3.8. Let X be a toric Deligne-Mumford stack with Deligne-Mumford torus T T BG. The morphism j : G X I gen X is an isomorphism. ! Proof. As the stack X is separated, we have that the natural morphism I X X is proper. As the projection G X X is a proper morphism, the morphism j is ! also a proper morphism. Its image contains ! the substack I T I gen T which is open and dense in I gen X . We deduce that the morphism j is birational. As the morphism j is e´tale, it is quasi-finite (cf. [19], Expose´ I, §3). The morphism j is proper hence closed and as its image contains the open dense torus, j is surjective. The morphism j is a representable, birational, surjective and quasi-finite morphism to the smooth Deligne-Mumford stack X. The stacky Zariski s main theorem C.1 finishes the proof. r
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
214 Fantechi, Mann and Nironi, Deligne-Mumford stacks
4. Canonical toric Deligne-Mumford stacks
In §4.1 we define the canonical smooth Deligne-Mumford stack associated to a variety with finite quotient singularities and we show that a canonical smooth Deligne- Mumford stack satisfies a universal property (Theorem 4.6). This should be well known, but we include it for the reader s convenience.
In §4.2, we characterize the canonical toric Deligne-Mumford stack via its coarse moduli space.
4.1. Canonical smooth Deligne-Mumford stacks. In this subsection, we do not as- sume that smooth Deligne-Mumford stacks are toric. First, we define canonical smooth Deligne-Mumford stacks and then we prove their universal property.
We recall a classical result.
Lemma 4.1. Let S be a smooth variety, and T be an a‰ne scheme. Let S 0 H Sbean open subvariety such that the complement has codimension at least 2 in S. Let f : S 0 Tbe a morphism. Then the morphism f extends uniquely to a morphism S T. ! ! Proof. The morphism f corresponds to an algebra homomorphism
K T G S 0; OS : ! 0 G O G O Since the complement has codimension 2, the restriction map S; S S 0; S 0 is an isomorphism. r !
Definition 4.2. (1) A dominant morphism f : V W of irreducible varieties is called codimension preserving if, for any irreducible closed! subvariety Z of W and every 1 irreducible component ZV of f Z , one has codimV ZV codimW Z. (2) A dominant morphism of orbifolds is called codimension preserving if the induced morphism on every irreducible component of the coarse moduli spaces is codimension preserving.
Remark 4.3. For any Deligne-Mumford stack, the structure morphism to the coarse moduli space is codimension preserving. Every flat morphism and in particular every smooth and e´tale morphism is codimension preserving. A composition of codimension preserving morphisms is codimension preserving.
Definition 4.4. Let X be an irreducible d-dimensional smooth Deligne-Mumford stack. Let e : X X be the structure morphism to the coarse moduli space. The Deligne- Mumford stack !X will be called canonical if the locus where e is not an isomorphism has dimension e d 2. Remark 4.5. Let X be a smooth canonical stack
(1) The locus where the structure map to the coarse moduli space e : X X is an isomorphism is precisely e 1 X , where X is the smooth locus of X. ! sm sm
AUTHOR’S COPY | AUTORENEXEMPLAR AUTHOR’S COPY | AUTORENEXEMPLAR
Fantechi, Mann and Nironi, Deligne-Mumford stacks 215
(2) The composition of the following isomorphisms
1 F 1 F F 1 F A X A X Pic X Pic e X Pic X ! sm ! sm ! sm ! 1 is the map sending D to O e D . Theorem 4.6 (universal property of canonical smooth Deligne-Mumford stacks). Let Y be a canonical smooth Deligne-Mumford stack, e : Y Y the morphism to the coarse moduli space, and f : X Y a dominant codimension preserving! morphism with X an orbi- fold. Then there exists a! unique, up to a unique 2-arrow, g : X Y such that the following
diagram is commutative: !
X b !g Y
! e f ? ? Y?: y Proof. We first prove uniqueness. Any two morphisms g, g making the diagram 1 1 commute must agree on the open dense subscheme f Ysm . Put i : f Ysm , X. Since Y is assumed to be separated, by Proposition 1.2, there exists a unique a : g !g such that ! a idi id. By uniqueness, it is enough to prove the result e´tale locally in Y, so we can assume that Y V=G where V is a smooth a‰ne variety and G a finite group acting on V with- out pseudo-reflections. It is enough to show that there exists an e´tale surjective morphism p : U X with U a smooth variety and a morphism g : U Y such that f p e g. In fact,! g is defined from g by descent, with the appropriate! compatibility conditions being taken care of by the uniqueness part. In this case Y V=G, and Y0 : V0=G where 1 V H V is the open locus where G acts freely. Let U : f p Y .As V =G is 0 0 0 0 isomorphic to Y0, we have a natural morphism U0 V0=G . This morphism defines a principal G-bundle P on U and a G-equivariant morphism! s : P V .
0 0 0 0 ! 0 s
P 0 V